What are the current challenges and limitations of path planning for cellular connected UAVs, and how can heuristic algorithms and reinforcement learning help to address these challenges?
Different from traditional path planning, cellular-connected UAV path planning has a more complex and dynamic situation, which may make the operation time longer. Most recent studies only focus on using traditional methods, such as dynamic programming, to solve the TSP problem. Only a few studies have used machine learning methods to address TSP. Additionally, most traditional algorithms can only handle simple TSPs that calculate the tour length between two cities using Euclidean distance, or that require a distance matrix for a sequence of cities. However, for cellular-connected UAVs, distance calculation is not a straightforward problem and cannot be calculated using Euclidean distance, so traditional algorithms may not be effective in new scenarios.
We consider a scenario where a UAV take off from a start state and fly over several places covered by cellular networks to finish some mission such as data collection or search and rescue. Once the missions are completed, UAV must fly back to where it takes off. In this paper, we use grid map to model the UAV flyable area and A* to calculate the distance between point and point. Then, we use three optimal algorithm, Genetic Algorithm (GA), Simulated Annealing Algorithm (SAA), Pointer Network (Reinforcement Learning) to calculate the best path between the star point and end point.
